Helping customers is essential for customer loyalty but it is complex & expensive. Through increased customer standards, desires and situations, providing customer support has become a tough challenge for service-driven organizations. 

Problem
How to keep the cost low on one hand and quality high on the other when human agents. 

Solution

Because this client already had a fully-operational virtual assistance, our focus was on automating the administrative part. We therefore built the Automated subject classification model and Automated summarizing model. Beroth models we able to successfully categorize journey & topic and summarize the chat conversation with over 80% accuracy.
Better insights: The automated models are on average 50-60% better in assigning the correct label & summary compared with a human agent.
Better decisions: With an 80% accuracy, the insights are also much more reliable for the management to substantiate their decisions.

Data

Half million ammonized WhatsApp messages

Technology

Unsupervised NLP model (Autoencoder & Transformers)

Scope

modelling, testing and implementation (MS Azure)

Planning

Realized in 24 weeks

Curious to know how to serve your customers with maximum quality and minimum costs?